Hub discovery in partial correlation graphical models

摘要: This paper treats the problem of screening a p-variate sample for strongly and multiply connected vertices in partial correlation graph associated with matrix sample. problem, called hub screening, is important many applications ranging from network security to computational biology finance social networks. In area security, node that becomes high neighboring nodes might signal anomalous activity such as coordinated flooding attack. set hubs gene expression can serve potential targets drug treatment block pathway or modulate host response. indicate vulnerable financial instrument sector whose collapse have major repercussions on market. networks observed interactions between criminal suspects could be an influential ringleader. The techniques theory presented this permit scalable reliable hubs. extends our previous work [arXiv:1102.1204] more challenging variables degree connectivity. particular we consider 1) extension difficult correlations exceeding specified magnitude; 2) vertex graph, often concentration exceeds degree.